A group based network analysis for Alzheimer's disease fMRI data

Network modeling are widely using in resting-state functional magnetic resonance imaging (rs-fMRI) for Alzheimer's disease (AD) research. Typically, Pearson correlation coefficient (PCC) was widely applied to construct brain connectivity network from BOLD signals of regions of interest. However, it often results in significant intra-group variability and complicates the identification of disease-specific functional connectivity patterns. To address this issue, we propose a novel brain network construction strategy, called SNBG, which uses aggregated information from the control group to derive a single-sample network. We compare SNBG and the PCC based method on a dataset from an Alzheimer's Disease Neuroimaging Initiative (ADNI) study. SNBG method captures more stable connections between regions of interest (ROIs) and increases classification accuracy from 89.24% of PCC based method to 97.13%. In addition, in AD-related local networks, such as default mode network (DMN), medial frontal network (MFN) and frontoparietal network (FPN), SNBG demonstrates lower intra-group heterogeneity than the PCC based method.